What is Background Remover?

Background Remover uses AI running entirely on your device to detect and remove image backgrounds in seconds. Upload a photo and get a clean transparent PNG — perfect for product photos, profile pictures, or design assets. No upload to any server, your images stay private.

The model running here is a U²-Net variant exported to ONNX, executed by ONNX Runtime Web with WebAssembly and (where supported) WebGPU acceleration. It produces a per-pixel alpha mask, then composites your subject onto transparency at the original resolution up to 2048 px on the longest side. Large photos are downscaled before inference; very long edges or fine hair benefit from the foreground preview before download.

How to use

1. Upload a photo (JPG, PNG, WebP, or HEIC from iPhone) by dropping it or clicking the upload area.
2. The AI model processes the image locally and separates the foreground from the background.
3. Preview the result with a checkerboard transparency pattern, then download the transparent PNG.

When to use

• Cleaning up product photos for a Shopify or Etsy listing so the catalogue feels consistent.
• Cutting yourself out of a busy snapshot for a LinkedIn or Slack avatar.
• Preparing transparent assets for design work in Figma, Canva, or Keynote.

Result

You need a headshot with a transparent background for your LinkedIn profile. Upload your photo, and within seconds the AI cleanly separates you from the cluttered office background — ready to place on any backdrop.

FAQ

Why does the AI struggle with hair or fur edges?

Hair has hundreds of semi-transparent strands and the model only outputs a single alpha value per pixel. You'll see a slight halo on flyaway hairs against a contrasting backdrop. Subjects shot against a plain, well-lit background give cleaner edges than those in a busy room.

What's the maximum resolution it supports?

Images are processed up to 2048 px on the longest side. Anything larger is automatically resampled to fit before inference, then the resulting mask is composited at the resized resolution. For higher resolutions, use a desktop image editor afterwards.

Does the image leave my device during processing?

No. The ONNX model and runtime are loaded into your browser memory on first use and inference runs locally. Nothing about your photo is sent over the network — even the model download is cached so subsequent runs work offline.

Why was the model load slow the first time?

The U²-Net weights are around 170 MB and have to download once before inference can start. Your browser caches them, so the next visit skips the download and the model loads in a couple of seconds.

Can I replace the removed background with a colour or another image?

The download is a transparent PNG, so any editor such as Photoshop, GIMP, Figma, or Pages can drop it onto a coloured layer or new backdrop. The preview here also offers a white-background view if you want to check how the cutout looks on light surfaces.

Related Tools